AgentOptimizer

[skzhang1]

Why are these changes needed?

In traditional ML pipeline, we train a model by updating its parameter according to the loss on the training set, while in the era of LLM agents, how should we train an agent? Here, we take an initial step towards the agent training. Inspired by the function calling capabilities provided by OpenAI, we draw an analogy between model parameters and agent functions/skills, and update agent’s functions/skills based on its historical performance on the training set.

In this notebook, we introduce a new class, ‘AgentOptimizer’, which is able to improve the function list of one Assistant-UserProxy pair according to the historical conversation histories. This feature would support agents in improving their ability to solve problems of the same type as previous tasks.

Any suggestions are appreciated.

Related issue number

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://microsoft.github.io/autogen/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (9b7a42f) 30.23% compared to head (a0870bd) 30.23%.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1053   +/-   ##
=======================================
  Coverage   30.23%   30.23%           
=======================================
  Files          30       30           
  Lines        3989     3989           
  Branches      899      899           
=======================================
  Hits         1206     1206           
  Misses       2704     2704           
  Partials       79       79           

Flag	Coverage Δ
unittests	30.18% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rickyloynd-microsoft]

@skzhang1 Thanks for the contribution. To put the PR in perspective, can you a general outline of techniques anticipated for updating the functions?

[skzhang1]

@skzhang1 Thanks for the contribution. To put the PR in perspective, can you a general outline of techniques anticipated for updating the functions?

@rickyloynd-microsoft Thanks for the suggestion. We just write a blog which provides an overview for this updating.